I wrote an essay a few weeks ago arguing that SaaS has hijacked Generative Engine Optimization (GEO), the same way agencies took over SEO a decade earlier. As a result, solo writers and small publishers have been left out, which is unfortunate, because the research says they are the ones GEO benefits most. 

After doing my due diligence, I decided to write this guide to help my fellow writers understand how GEO works and how they can use it to get cited by different AI engines like ChatGPT, Gemini, and Perplexity.

Before any of the tactics make sense, you need a working picture of what happens when someone asks ChatGPT or Perplexity a question.

When a user types a question into an AI search tool, the model does not look up a single page the way Google used to do. It uses a process called Retrieval-Augmented Generation or RAG. And RAG works in three steps. 

As a result, two things happen:

The first is that AI search engines do not cite pages, they cite passages. A 3,000-word essay might have one paragraph extracted and cited, and the other 95% of it ignored. This means the unit of optimization in GEO is not the article but the information presented in that article. The Princeton research was the first study to measure it systematically and then almost every other study confirmed this pattern.

The second is that AI systems read content the way an editor skims a piece looking for the quotable line. The systems scan for self-contained claims with clear attribution and reward declarative writing. An article or blog with hedging content simply does not get cited. 

An important thing to remember about AI citations is that the system extracts specifically and cites narrowly. So, it doesn't matter how impressive a piece you have written, if it doesn't contain a clear answer to a query, the AI has nothing to cite. The content could be excellent, yet still invisible to AI.

That is all GEO is, at the mechanical level. The rest of this handbook is about how to write so you can get cited by AI.

In 2024, researchers at Princeton and IIT Delhi published a paper titled GEO: Generative Engine Optimization, presented at the KDD conference, that tested 9 content optimization tactics across 10,000 queries on a system designed to mimic Bing Chat. 

5 of the 9 tactics produced meaningful gains in AI citation rates. 4 either did nothing or affected the performance negatively. The 5 techniques that worked are the foundation of every legitimate piece of GEO advice in circulation, and they are the foundation of this handbook.

Here are 5 tips you can use to GEO cited by AI.

Include inline references to credible external sources for every factual claim you make. These references must be added in the body of the text next to the claim with the source named. Do not add the sources at the bottom of your article/blog.

Why it works: AI systems use external citations as a trust signal. A passage that says "according to Pew Research, 41% of Gen Z now use social media as their first point of search" gives the AI two useful things at once. 

It not only gives a verifiable claim but also a credible source the AI can cross-reference. A passage that just says "most young people use social media to search" gives the AI nothing and thus isn't citeable.

Effect: The Princeton paper found that Cite Sources produced a 30% to 40% lift in AI citation rates across the queries tested. For low-ranking pages specifically, the lift was 115%. This is the single highest-leverage tactic in the entire study for small publishers.

How to do it: When you make a factual claim, name the source in the sentence and link to the primary source, not a secondary article quoting it. "Pew Research found" beats "research has found." 

Add a link to the Pew report itself, not to a Forbes article summarizing the Pew report. Aim for at least one cited source per substantive section. Government sites, academic papers, named studies and established publications all work. Avoid linking to sources that are promotional, like company blogs citing their own data without a methodology link.

Before:

Research has shown that AI search is growing rapidly.

After:

According to Previsible's 2025 report, AI-referred web sessions grew 527% year on year between January and May 2025.

Always replace vague claims with specific numerical data. For example, "most" becomes 73%. "Recently" becomes "in March 2026." "A lot of growth" becomes "from 400 million weekly users to 800 million in 8 months."

Why it works: AI systems extract data points more reliably than narrative claims. A passage with specific numbers is computationally easier for the model to cite, because the number serves as a clean anchor for the attribution. 

Without specific numbers, the model has nothing clean to lift. It either ends up paraphrasing, which reduces citation likelihood, or skipping the passage entirely.

Effect: Adding statistics to content produced a 30% to 41% lift in the Princeton study, the strongest single effect of any tactic tested. The effect was largest in fact-driven domains like law, government and opinionated content.

How to do it: Every time you write a vague quantifier, ask yourself if you can replace it with a specific number. If you can find the real number in 60 seconds of searching, find it. If you cannot, consider whether the claim is worth adding at all. 

One important caution: the statistic needs to be real and sourced. Stuffing random numbers suh as, such as claiming "70% of users prefer our product" without a source, produces a citation drop, not a lift, because the model cannot verify the number and the surrounding prose looks artificial.

Before:

AI search has grown enormously in the past year.

After:

AI search has grown 527% year on year between January and May 2025, according to Previsible, and ChatGPT alone now serves roughly 800 million weekly users.

Try adding a quote from a recognizable expert or authority, attributed by name, with their credentials in your content.

Why it work: Quotation marks around a named expert function as a credibility signal that AI models heavily reward. The quotation gives the model a self-contained, attributable claim it can lift without rewording, and the named expert gives it a way to anchor the attribution. AI systems extract quoted material at much higher rates than paraphrased content.

Effect: Adding a quotation to content produced a 30% to 41% lift in the Princeton study, roughly tied with Statistics Addition as the highest-impact technique to get GEO cited. One later analysis found that quoted material increased AI visibility by 37% specifically.

How to do it. When you reference someone's argument or finding, quote them directly rather than paraphrasing. Use their full name and a relevant credential the first time they appear. If you cannot quote the source you are referencing, find a different source you can quote. 

For essays where direct expert quotes are hard to come by, quoting from a specific page of a named study or report works almost as well.

Before:

Experts say AI search is changing how people find information.

After:

"There is no such thing as GEO or AEO without SEO fundamentals," Google's John Mueller said in January 2026, a remark that captured the strange status of AI search optimization at the moment of its emergence.

The prose in your content should be clean, grammatically correct, and natural to read. It does not need to be elegant or distinctive; it simply needs to read as though a capable writer sat down and wrote it. This means no stumbling syntax, no bloated phrasing and no sentences that make the reader pause because they’re trying to comprehend the paragraph.

Why it works: This sounds obvious but it works. AI models evaluate the readability of a passage when deciding whether to extract from it, because passages that read poorly are harder to cite without making the AI's own response read poorly. A well-written passage requires less rewriting from the model, which means the model cites it more readily.

Effect: Fluency optimization produced a meaningful citation lift in the Princeton study, though the effect was smaller than Cite Sources, Statistics, and Quotations. It also interacts with the others. A well-written passage that contains a statistic and a citation outperforms a poorly written one that contains the same.

How to do it. Read your draft aloud and cut anything that makes you stumble. Remove filler words like "actually," "really," "very," "essentially," and "basically." Replace abstract nouns with concrete ones. Break long sentences into shorter ones, but not into so short that they feel chopped. Use specific verbs and if a sentence sounds like it came from a press release, rewrite it.

Before:

It is increasingly the case that there has been a really significant shift in the way that people are essentially using AI tools to actually find information online.

After:

People have started using AI tools to find information online, and the shift has been fast.

Try writing with authority. Authoritative voice means writing in declarative sentences without hedging every claim. It does not mean overstating what you know.

Why it works: AI systems cite declarative statements at much higher rates than hedged ones. One analysis found that declarative phrasing ("X is defined as") was cited 36.2% of the time, while hedged phrasing ("X can sometimes be thought of as") was cited 20.2% of the time. The model treats a confident sentence as a citable claim and a hedged one as background information.

Effect size: Authoritative writing produced a meaningful citation lift in the said Princeton study, smaller than the top three tactics but consistent across domains. The effect was largest in opinion writing and analysis.

How to do it. Drop "I think," "perhaps," "it seems that," "one could argue," and similar phrases unless you genuinely mean to flag uncertainty. Replace "this might suggest" with "this suggests." Replace "some experts believe" with the specific expert's name and their belief. If you do not feel confident enough to make a claim plainly, the right answer is usually to do more research, not to soften the claim.

Before:

It seems like maybe one possible interpretation is that GEO might be more accessible to small writers than is commonly thought.

After:

GEO is more accessible to small writers than the current discourse suggests and the research supports this.

Here are four old tricks that don’t work with GEO. 

Of everything in this handbook, the answer capsule is one of the most significant tools you can use to get cited by AI. It produces immediate impact, especially on essays you have already published and only takes about 10 minutes per essay.

An answer capsule is a 40 to 60-word self-contained direct answer placed immediately below an H2 heading, before any other text in the section. It answers the question the H2 implies, in plain declarative prose, without links or hedging. It is written to be lifted verbatim by an AI system.

The research on this is unusually clean. One analysis of 2 million ChatGPT sessions found that 72.4%  cited blog posts contained an identifiable answer capsule. Separately, research from Kevin Indig found that 44.2% of all LLM citations come from the first 30% of an article's text. 

These two findings reinforce each other and highlight that AI systems extract heavily from the opening of each section. An AI model extracts most reliably when the opening of a section is a self-contained answer.

Here is an example for your reference:

Before:

Why solo writers benefit from GEO

There are a lot of reasons why solo writers might want to think about GEO, and we will get into them, but first it is worth understanding a bit of context about how AI search works and why this matters now.

After:

Why solo writers benefit from GEO

Solo writers benefit from GEO because the research shows lower-ranking pages get a 115% visibility lift from citing sources properly, a much larger effect than the same tactic produces on established sites. Domain authority and backlinks, the structural advantages big publishers spent decades building, do not carry over to AI search nearly as cleanly.

Did you notice the difference? The "after" version contains the answer, the statistic, the source attribution, and a clean supporting claim, all in the first 50 words. An AI system can lift it as a complete unit. 

And when you compare it with the "before" version, you see it contains no extractable claim, only a promise that one is coming. And in such instances, AI has nothing to lift for citation.

Two practical notes on capsules. 

First, do not put links inside the capsule itself. Because links suggest to the model that the real answer is somewhere else and it ends up reducing the capsule's perceived self-sufficiency. Put your links in the paragraph immediately after. 

Second, write the capsule last. Draft the section first, once you have completed your piece, then go back and compress the answer into the opening 40 to 60 words.

Beyond the answer capsule, the structure of a piece as a whole affects how often AI systems cite from it. The patterns are well-documented but rarely explained outside of agency content. 

Here’s how you can structure your blog/article to get GEO cited.

Each section under an H2 should be roughly 120 to 180 words. If it's shorter than 120 and the section lacks enough context to stand alone as a citable unit, it gets removed. If a section is longer than 180, the AI may extract a fragment that misses the point. The "information island" test applies to every section: if it were extracted on its own, would it still make sense?

AI systems match headings to user queries. An H2 written as a question that a person would ask gets extracted at higher rates than an H2 written as a topic label. "How do solo writers get cited by ChatGPT" beats "Solo writers and ChatGPT citation." Not every H2 needs to be a question, but the ones covering your main claims should be.

This point cannot be stressed enough; even outside the formal answer capsule format, the first sentence of each section should answer the question the section is about. Background, context, and setup come second. AI reads opening of each section closely and the rest more loosely.

The single most important finding for structure is that 44.2% of AI citations come from the first 30% of a piece's text. Whatever your strongest, most citable claim is, it should appear in the first third, not in the conclusion. The conclusion is where humans look, AI simply looks at the opening.

Do not write sections that depend on context from earlier sections. The AI may extract any section without reading the others. Each section should make sense as a standalone, even if that creates some mild repetition.

I wrote this guide for writers who are trying to figure out GEO on their own without an agency budget or a SaaS subscription. I hope it gives you something you can use to your advantage.